Abstract

Soil organic carbon (SOC) is a critical component of soil health, influencing agricultural productivity, carbon sequestration, and climate change mitigation. Accurate SOC prediction over large areas is challenging due to spatial variability and the limitations of traditional soil sampling. This article investigates the use of ensemble machine learning models, integrating remote sensing data, to predict SOC content across a 500-hectare agricultural region in Saskatchewan, Canada. Multispectral satellite imagery from Sentinel-2, combined with topographic and climatic data, was used to train ensemble models, including Random Forest, Gradient Boosting, and XGBoost. The models achieved an average R² of 0.89 and a root mean square error (RMSE) of 0.31% for SOC prediction. Results demonstrate the superiority of ensemble methods over single-model approaches, with Random Forest outperforming others in accuracy and robustness. Challenges such as data resolution and model interpretability are discussed, alongside future directions for integrating hyperspectral data and deep learning.